Loop unroll; avoid bounds checks

[benaadams]

No description provided.

[davidfowl]

@benaadams Before after numbers?

[benaadams]

Will benchmark

[davidfowl]

?

[benaadams]

I got it wrong, is count not upper bound. Though current code still pass with this change.

[benaadams]

Test cases https://gist.github.com/benaadams/079a1ded2e14bc243fd5

[benaadams]

Current is SequentialArray proposed is UnrolledParallelPointer

BenchmarkDotNet=v0.7.8.0
OS=Microsoft Windows NT 6.2.9200.0
Processor=Intel(R) Core(TM) i7-4720HQ CPU @ 2.60GHz, ProcessorCount=8
HostCLR=MS.NET 4.0.30319.42000, Arch=64-bit  [RyuJIT]
Type=AsciiToString  Mode=Throughput  Platform=X64  Jit=RyuJit  .NET=HostFramework

Method	Param	AvrTime	StdDev	op/s
SequentialArray	1	1.8446 ns	0.0359 ns	542,130,112.58
SequentialPointer	1	1.4709 ns	0.0290 ns	679,872,918.30
UnrolledPointer	1	1.5475 ns	0.0068 ns	646,184,034.29
UnrolledParallelPointer	1	1.4456 ns	0.0280 ns	691,731,058.85
SequentialArray	2	1.8347 ns	0.0301 ns	545,041,308.79
SequentialPointer	2	1.7573 ns	0.0055 ns	569,058,870.74
UnrolledPointer	2	1.7582 ns	0.0302 ns	568,748,690.53
UnrolledParallelPointer	2	1.6387 ns	0.0255 ns	610,257,762.49
SequentialArray	8	2.6034 ns	0.0473 ns	384,119,102.12
SequentialPointer	8	2.7821 ns	0.0495 ns	359,435,697.96
UnrolledPointer	8	2.3780 ns	0.0419 ns	420,515,824.62
UnrolledParallelPointer	8	1.9083 ns	0.0923 ns	524,037,000.21
SequentialArray	64	15.5845 ns	1.2281 ns	64,166,181.67
SequentialPointer	64	12.2456 ns	0.2320 ns	81,662,055.86
UnrolledPointer	64	9.5331 ns	0.1572 ns	104,897,218.65
UnrolledParallelPointer	64	5.8002 ns	0.1617 ns	172,408,326.49
SequentialArray	512	109.1622 ns	3.9246 ns	9,160,708.35
SequentialPointer	512	77.9059 ns	1.3984 ns	12,836,002.73
UnrolledPointer	512	68.6465 ns	1.2514 ns	14,567,379.93
UnrolledParallelPointer	512	39.9602 ns	1.6196 ns	25,024,919.12
SequentialArray	1024	216.8628 ns	4.3980 ns	4,611,210.56
SequentialPointer	1024	147.6454 ns	2.6145 ns	6,772,985.46
UnrolledPointer	1024	134.3055 ns	2.3314 ns	7,445,713.61
UnrolledParallelPointer	1024	77.5874 ns	1.8344 ns	12,888,690.04
SequentialArray	4096	838.0113 ns	8.0319 ns	1,193,301.35
SequentialPointer	4096	603.3346 ns	10.7317 ns	1,657,455.03
UnrolledPointer	4096	528.5438 ns	8.8499 ns	1,891,990.90
UnrolledParallelPointer	4096	306.2025 ns	4.9520 ns	3,265,812.70
SequentialArray	8192	1,665.9055 ns	17.9379 ns	600,274.20
SequentialPointer	8192	1,203.2110 ns	20.8846 ns	831,109.43
UnrolledPointer	8192	1,054.4902 ns	0.7537 ns	948,325.53
UnrolledParallelPointer	8192	606.6170 ns	10.3170 ns	1,648,486.66
SequentialArray	16384	3,331.0892 ns	31.2668 ns	300,202.12
SequentialPointer	16384	2,404.2652 ns	1.2482 ns	415,927.49
UnrolledPointer	16384	2,105.9399 ns	2.7216 ns	474,847.36
UnrolledParallelPointer	16384	1,447.2493 ns	25.6027 ns	690,965.98

Approx 1 hour to benchmark

[benaadams]

@davidfowl new implementation is

x 1.36 for 8 bytes
x 2 for 64 bytes
x 2.7 for 512 bytes

(not including the string constructor which is the same in all cases; and the GC cost would dwarf the ASCII decode)

[benaadams]

Will come back to this